Conversion/SPIRVToLLVM/SPIRVToLLVM.cpp

//===- SPIRVToLLVM.cpp - SPIR-V to LLVM Patterns --------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements patterns to convert SPIR-V dialect to LLVM dialect.
//
//===----------------------------------------------------------------------===//

#include "mlir/Conversion/SPIRVToLLVM/SPIRVToLLVM.h"
#include "mlir/Conversion/StandardToLLVM/ConvertStandardToLLVM.h"
#include "mlir/Conversion/StandardToLLVM/ConvertStandardToLLVMPass.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
#include "mlir/Dialect/SPIRV/IR/SPIRVDialect.h"
#include "mlir/Dialect/SPIRV/IR/SPIRVOps.h"
#include "mlir/Dialect/SPIRV/Utils/LayoutUtils.h"
#include "mlir/Dialect/StandardOps/IR/Ops.h"
#include "mlir/IR/BuiltinOps.h"
#include "mlir/IR/PatternMatch.h"
#include "mlir/Support/LogicalResult.h"
#include "mlir/Transforms/DialectConversion.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/FormatVariadic.h"

#define DEBUG_TYPE "spirv-to-llvm-pattern"

using namespace mlir;

//===----------------------------------------------------------------------===//
// Utility functions
//===----------------------------------------------------------------------===//

/// Returns true if the given type is a signed integer or vector type.
static bool isSignedIntegerOrVector(Type type) {
  if (type.isSignedInteger())
    return true;
  if (auto vecType = type.dyn_cast<VectorType>())
    return vecType.getElementType().isSignedInteger();
  return false;
}

/// Returns true if the given type is an unsigned integer or vector type
static bool isUnsignedIntegerOrVector(Type type) {
  if (type.isUnsignedInteger())
    return true;
  if (auto vecType = type.dyn_cast<VectorType>())
    return vecType.getElementType().isUnsignedInteger();
  return false;
}

/// Returns the bit width of integer, float or vector of float or integer values
static unsigned getBitWidth(Type type) {
  assert((type.isIntOrFloat() || type.isa<VectorType>()) &&
         "bitwidth is not supported for this type");
  if (type.isIntOrFloat())
    return type.getIntOrFloatBitWidth();
  auto vecType = type.dyn_cast<VectorType>();
  auto elementType = vecType.getElementType();
  assert(elementType.isIntOrFloat() &&
         "only integers and floats have a bitwidth");
  return elementType.getIntOrFloatBitWidth();
}

/// Returns the bit width of LLVMType integer or vector.
static unsigned getLLVMTypeBitWidth(Type type) {
  return (LLVM::isCompatibleVectorType(type) ? LLVM::getVectorElementType(type)
                                             : type)
      .cast<IntegerType>()
      .getWidth();
}

/// Creates `IntegerAttribute` with all bits set for given type
static IntegerAttr minusOneIntegerAttribute(Type type, Builder builder) {
  if (auto vecType = type.dyn_cast<VectorType>()) {
    auto integerType = vecType.getElementType().cast<IntegerType>();
    return builder.getIntegerAttr(integerType, -1);
  }
  auto integerType = type.cast<IntegerType>();
  return builder.getIntegerAttr(integerType, -1);
}

/// Creates `llvm.mlir.constant` with all bits set for the given type.
static Value createConstantAllBitsSet(Location loc, Type srcType, Type dstType,
                                      PatternRewriter &rewriter) {
  if (srcType.isa<VectorType>()) {
    return rewriter.create<LLVM::ConstantOp>(
        loc, dstType,
        SplatElementsAttr::get(srcType.cast<ShapedType>(),
                               minusOneIntegerAttribute(srcType, rewriter)));
  }
  return rewriter.create<LLVM::ConstantOp>(
      loc, dstType, minusOneIntegerAttribute(srcType, rewriter));
}

/// Creates `llvm.mlir.constant` with a floating-point scalar or vector value.
static Value createFPConstant(Location loc, Type srcType, Type dstType,
                              PatternRewriter &rewriter, double value) {
  if (auto vecType = srcType.dyn_cast<VectorType>()) {
    auto floatType = vecType.getElementType().cast<FloatType>();
    return rewriter.create<LLVM::ConstantOp>(
        loc, dstType,
        SplatElementsAttr::get(vecType,
                               rewriter.getFloatAttr(floatType, value)));
  }
  auto floatType = srcType.cast<FloatType>();
  return rewriter.create<LLVM::ConstantOp>(
      loc, dstType, rewriter.getFloatAttr(floatType, value));
}

/// Utility function for bitfield ops:
///   - `BitFieldInsert`
///   - `BitFieldSExtract`
///   - `BitFieldUExtract`
/// Truncates or extends the value. If the bitwidth of the value is the same as
/// `llvmType` bitwidth, the value remains unchanged.
static Value optionallyTruncateOrExtend(Location loc, Value value,
                                        Type llvmType,
                                        PatternRewriter &rewriter) {
  auto srcType = value.getType();
  unsigned targetBitWidth = getLLVMTypeBitWidth(llvmType);
  unsigned valueBitWidth = LLVM::isCompatibleType(srcType)
                               ? getLLVMTypeBitWidth(srcType)
                               : getBitWidth(srcType);

  if (valueBitWidth < targetBitWidth)
    return rewriter.create<LLVM::ZExtOp>(loc, llvmType, value);
  // If the bit widths of `Count` and `Offset` are greater than the bit width
  // of the target type, they are truncated. Truncation is safe since `Count`
  // and `Offset` must be no more than 64 for op behaviour to be defined. Hence,
  // both values can be expressed in 8 bits.
  if (valueBitWidth > targetBitWidth)
    return rewriter.create<LLVM::TruncOp>(loc, llvmType, value);
  return value;
}

/// Broadcasts the value to vector with `numElements` number of elements.
static Value broadcast(Location loc, Value toBroadcast, unsigned numElements,
                       LLVMTypeConverter &typeConverter,
                       ConversionPatternRewriter &rewriter) {
  auto vectorType = VectorType::get(numElements, toBroadcast.getType());
  auto llvmVectorType = typeConverter.convertType(vectorType);
  auto llvmI32Type = typeConverter.convertType(rewriter.getIntegerType(32));
  Value broadcasted = rewriter.create<LLVM::UndefOp>(loc, llvmVectorType);
  for (unsigned i = 0; i < numElements; ++i) {
    auto index = rewriter.create<LLVM::ConstantOp>(
        loc, llvmI32Type, rewriter.getI32IntegerAttr(i));
    broadcasted = rewriter.create<LLVM::InsertElementOp>(
        loc, llvmVectorType, broadcasted, toBroadcast, index);
  }
  return broadcasted;
}

/// Broadcasts the value. If `srcType` is a scalar, the value remains unchanged.
static Value optionallyBroadcast(Location loc, Value value, Type srcType,
                                 LLVMTypeConverter &typeConverter,
                                 ConversionPatternRewriter &rewriter) {
  if (auto vectorType = srcType.dyn_cast<VectorType>()) {
    unsigned numElements = vectorType.getNumElements();
    return broadcast(loc, value, numElements, typeConverter, rewriter);
  }
  return value;
}

/// Utility function for bitfield ops: `BitFieldInsert`, `BitFieldSExtract` and
/// `BitFieldUExtract`.
/// Broadcast `Offset` and `Count` to match the type of `Base`. If `Base` is of
/// a vector type, construct a vector that has:
///  - same number of elements as `Base`
///  - each element has the type that is the same as the type of `Offset` or
///    `Count`
///  - each element has the same value as `Offset` or `Count`
/// Then cast `Offset` and `Count` if their bit width is different
/// from `Base` bit width.
static Value processCountOrOffset(Location loc, Value value, Type srcType,
                                  Type dstType, LLVMTypeConverter &converter,
                                  ConversionPatternRewriter &rewriter) {
  Value broadcasted =
      optionallyBroadcast(loc, value, srcType, converter, rewriter);
  return optionallyTruncateOrExtend(loc, broadcasted, dstType, rewriter);
}

/// Converts SPIR-V struct with a regular (according to `VulkanLayoutUtils`)
/// offset to LLVM struct. Otherwise, the conversion is not supported.
static Optional<Type>
convertStructTypeWithOffset(spirv::StructType type,
                            LLVMTypeConverter &converter) {
  if (type != VulkanLayoutUtils::decorateType(type))
    return llvm::None;

  auto elementsVector = llvm::to_vector<8>(
      llvm::map_range(type.getElementTypes(), [&](Type elementType) {
        return converter.convertType(elementType);
      }));
  return LLVM::LLVMStructType::getLiteral(type.getContext(), elementsVector,
                                          /*isPacked=*/false);
}

/// Converts SPIR-V struct with no offset to packed LLVM struct.
static Type convertStructTypePacked(spirv::StructType type,
                                    LLVMTypeConverter &converter) {
  auto elementsVector = llvm::to_vector<8>(
      llvm::map_range(type.getElementTypes(), [&](Type elementType) {
        return converter.convertType(elementType);
      }));
  return LLVM::LLVMStructType::getLiteral(type.getContext(), elementsVector,
                                          /*isPacked=*/true);
}

/// Creates LLVM dialect constant with the given value.
static Value createI32ConstantOf(Location loc, PatternRewriter &rewriter,
                                 unsigned value) {
  return rewriter.create<LLVM::ConstantOp>(
      loc, IntegerType::get(rewriter.getContext(), 32),
      rewriter.getIntegerAttr(rewriter.getI32Type(), value));
}

/// Utility for `spv.Load` and `spv.Store` conversion.
static LogicalResult replaceWithLoadOrStore(Operation *op,
                                            ConversionPatternRewriter &rewriter,
                                            LLVMTypeConverter &typeConverter,
                                            unsigned alignment, bool isVolatile,
                                            bool isNonTemporal) {
  if (auto loadOp = dyn_cast<spirv::LoadOp>(op)) {
    auto dstType = typeConverter.convertType(loadOp.getType());
    if (!dstType)
      return failure();
    rewriter.replaceOpWithNewOp<LLVM::LoadOp>(
        loadOp, dstType, loadOp.ptr(), alignment, isVolatile, isNonTemporal);
    return success();
  }
  auto storeOp = cast<spirv::StoreOp>(op);
  rewriter.replaceOpWithNewOp<LLVM::StoreOp>(storeOp, storeOp.value(),
                                             storeOp.ptr(), alignment,
                                             isVolatile, isNonTemporal);
  return success();
}

//===----------------------------------------------------------------------===//
// Type conversion
//===----------------------------------------------------------------------===//

/// Converts SPIR-V array type to LLVM array. Natural stride (according to
/// `VulkanLayoutUtils`) is also mapped to LLVM array. This has to be respected
/// when converting ops that manipulate array types.
static Optional<Type> convertArrayType(spirv::ArrayType type,
                                       TypeConverter &converter) {
  unsigned stride = type.getArrayStride();
  Type elementType = type.getElementType();
  auto sizeInBytes = elementType.cast<spirv::SPIRVType>().getSizeInBytes();
  if (stride != 0 &&
      !(sizeInBytes.hasValue() && sizeInBytes.getValue() == stride))
    return llvm::None;

  auto llvmElementType = converter.convertType(elementType);
  unsigned numElements = type.getNumElements();
  return LLVM::LLVMArrayType::get(llvmElementType, numElements);
}

/// Converts SPIR-V pointer type to LLVM pointer. Pointer's storage class is not
/// modelled at the moment.
static Type convertPointerType(spirv::PointerType type,
                               TypeConverter &converter) {
  auto pointeeType = converter.convertType(type.getPointeeType());
  return LLVM::LLVMPointerType::get(pointeeType);
}

/// Converts SPIR-V runtime array to LLVM array. Since LLVM allows indexing over
/// the bounds, the runtime array is converted to a 0-sized LLVM array. There is
/// no modelling of array stride at the moment.
static Optional<Type> convertRuntimeArrayType(spirv::RuntimeArrayType type,
                                              TypeConverter &converter) {
  if (type.getArrayStride() != 0)
    return llvm::None;
  auto elementType = converter.convertType(type.getElementType());
  return LLVM::LLVMArrayType::get(elementType, 0);
}

/// Converts SPIR-V struct to LLVM struct. There is no support of structs with
/// member decorations. Also, only natural offset is supported.
static Optional<Type> convertStructType(spirv::StructType type,
                                        LLVMTypeConverter &converter) {
  SmallVector<spirv::StructType::MemberDecorationInfo, 4> memberDecorations;
  type.getMemberDecorations(memberDecorations);
  if (!memberDecorations.empty())
    return llvm::None;
  if (type.hasOffset())
    return convertStructTypeWithOffset(type, converter);
  return convertStructTypePacked(type, converter);
}

//===----------------------------------------------------------------------===//
// Operation conversion
//===----------------------------------------------------------------------===//

namespace {

class AccessChainPattern : public SPIRVToLLVMConversion<spirv::AccessChainOp> {
public:
  using SPIRVToLLVMConversion<spirv::AccessChainOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::AccessChainOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto dstType = typeConverter.convertType(op.component_ptr().getType());
    if (!dstType)
      return failure();
    // To use GEP we need to add a first 0 index to go through the pointer.
    auto indices = llvm::to_vector<4>(op.indices());
    Type indexType = op.indices().front().getType();
    auto llvmIndexType = typeConverter.convertType(indexType);
    if (!llvmIndexType)
      return failure();
    Value zero = rewriter.create<LLVM::ConstantOp>(
        op.getLoc(), llvmIndexType, rewriter.getIntegerAttr(indexType, 0));
    indices.insert(indices.begin(), zero);
    rewriter.replaceOpWithNewOp<LLVM::GEPOp>(op, dstType, op.base_ptr(),
                                             indices);
    return success();
  }
};

class AddressOfPattern : public SPIRVToLLVMConversion<spirv::AddressOfOp> {
public:
  using SPIRVToLLVMConversion<spirv::AddressOfOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::AddressOfOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto dstType = typeConverter.convertType(op.pointer().getType());
    if (!dstType)
      return failure();
    rewriter.replaceOpWithNewOp<LLVM::AddressOfOp>(op, dstType, op.variable());
    return success();
  }
};

class BitFieldInsertPattern
    : public SPIRVToLLVMConversion<spirv::BitFieldInsertOp> {
public:
  using SPIRVToLLVMConversion<spirv::BitFieldInsertOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::BitFieldInsertOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto srcType = op.getType();
    auto dstType = typeConverter.convertType(srcType);
    if (!dstType)
      return failure();
    Location loc = op.getLoc();

    // Process `Offset` and `Count`: broadcast and extend/truncate if needed.
    Value offset = processCountOrOffset(loc, op.offset(), srcType, dstType,
                                        typeConverter, rewriter);
    Value count = processCountOrOffset(loc, op.count(), srcType, dstType,
                                       typeConverter, rewriter);

    // Create a mask with bits set outside [Offset, Offset + Count - 1].
    Value minusOne = createConstantAllBitsSet(loc, srcType, dstType, rewriter);
    Value maskShiftedByCount =
        rewriter.create<LLVM::ShlOp>(loc, dstType, minusOne, count);
    Value negated = rewriter.create<LLVM::XOrOp>(loc, dstType,
                                                 maskShiftedByCount, minusOne);
    Value maskShiftedByCountAndOffset =
        rewriter.create<LLVM::ShlOp>(loc, dstType, negated, offset);
    Value mask = rewriter.create<LLVM::XOrOp>(
        loc, dstType, maskShiftedByCountAndOffset, minusOne);

    // Extract unchanged bits from the `Base`  that are outside of
    // [Offset, Offset + Count - 1]. Then `or` with shifted `Insert`.
    Value baseAndMask =
        rewriter.create<LLVM::AndOp>(loc, dstType, op.base(), mask);
    Value insertShiftedByOffset =
        rewriter.create<LLVM::ShlOp>(loc, dstType, op.insert(), offset);
    rewriter.replaceOpWithNewOp<LLVM::OrOp>(op, dstType, baseAndMask,
                                            insertShiftedByOffset);
    return success();
  }
};

/// Converts SPIR-V ConstantOp with scalar or vector type.
class ConstantScalarAndVectorPattern
    : public SPIRVToLLVMConversion<spirv::ConstantOp> {
public:
  using SPIRVToLLVMConversion<spirv::ConstantOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::ConstantOp constOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto srcType = constOp.getType();
    if (!srcType.isa<VectorType>() && !srcType.isIntOrFloat())
      return failure();

    auto dstType = typeConverter.convertType(srcType);
    if (!dstType)
      return failure();

    // SPIR-V constant can be a signed/unsigned integer, which has to be
    // casted to signless integer when converting to LLVM dialect. Removing the
    // sign bit may have unexpected behaviour. However, it is better to handle
    // it case-by-case, given that the purpose of the conversion is not to
    // cover all possible corner cases.
    if (isSignedIntegerOrVector(srcType) ||
        isUnsignedIntegerOrVector(srcType)) {
      auto signlessType = rewriter.getIntegerType(getBitWidth(srcType));

      if (srcType.isa<VectorType>()) {
        auto dstElementsAttr = constOp.value().cast<DenseIntElementsAttr>();
        rewriter.replaceOpWithNewOp<LLVM::ConstantOp>(
            constOp, dstType,
            dstElementsAttr.mapValues(
                signlessType, [&](const APInt &value) { return value; }));
        return success();
      }
      auto srcAttr = constOp.value().cast<IntegerAttr>();
      auto dstAttr = rewriter.getIntegerAttr(signlessType, srcAttr.getValue());
      rewriter.replaceOpWithNewOp<LLVM::ConstantOp>(constOp, dstType, dstAttr);
      return success();
    }
    rewriter.replaceOpWithNewOp<LLVM::ConstantOp>(constOp, dstType, operands,
                                                  constOp->getAttrs());
    return success();
  }
};

class BitFieldSExtractPattern
    : public SPIRVToLLVMConversion<spirv::BitFieldSExtractOp> {
public:
  using SPIRVToLLVMConversion<spirv::BitFieldSExtractOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::BitFieldSExtractOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto srcType = op.getType();
    auto dstType = typeConverter.convertType(srcType);
    if (!dstType)
      return failure();
    Location loc = op.getLoc();

    // Process `Offset` and `Count`: broadcast and extend/truncate if needed.
    Value offset = processCountOrOffset(loc, op.offset(), srcType, dstType,
                                        typeConverter, rewriter);
    Value count = processCountOrOffset(loc, op.count(), srcType, dstType,
                                       typeConverter, rewriter);

    // Create a constant that holds the size of the `Base`.
    IntegerType integerType;
    if (auto vecType = srcType.dyn_cast<VectorType>())
      integerType = vecType.getElementType().cast<IntegerType>();
    else
      integerType = srcType.cast<IntegerType>();

    auto baseSize = rewriter.getIntegerAttr(integerType, getBitWidth(srcType));
    Value size =
        srcType.isa<VectorType>()
            ? rewriter.create<LLVM::ConstantOp>(
                  loc, dstType,
                  SplatElementsAttr::get(srcType.cast<ShapedType>(), baseSize))
            : rewriter.create<LLVM::ConstantOp>(loc, dstType, baseSize);

    // Shift `Base` left by [sizeof(Base) - (Count + Offset)], so that the bit
    // at Offset + Count - 1 is the most significant bit now.
    Value countPlusOffset =
        rewriter.create<LLVM::AddOp>(loc, dstType, count, offset);
    Value amountToShiftLeft =
        rewriter.create<LLVM::SubOp>(loc, dstType, size, countPlusOffset);
    Value baseShiftedLeft = rewriter.create<LLVM::ShlOp>(
        loc, dstType, op.base(), amountToShiftLeft);

    // Shift the result right, filling the bits with the sign bit.
    Value amountToShiftRight =
        rewriter.create<LLVM::AddOp>(loc, dstType, offset, amountToShiftLeft);
    rewriter.replaceOpWithNewOp<LLVM::AShrOp>(op, dstType, baseShiftedLeft,
                                              amountToShiftRight);
    return success();
  }
};

class BitFieldUExtractPattern
    : public SPIRVToLLVMConversion<spirv::BitFieldUExtractOp> {
public:
  using SPIRVToLLVMConversion<spirv::BitFieldUExtractOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::BitFieldUExtractOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto srcType = op.getType();
    auto dstType = typeConverter.convertType(srcType);
    if (!dstType)
      return failure();
    Location loc = op.getLoc();

    // Process `Offset` and `Count`: broadcast and extend/truncate if needed.
    Value offset = processCountOrOffset(loc, op.offset(), srcType, dstType,
                                        typeConverter, rewriter);
    Value count = processCountOrOffset(loc, op.count(), srcType, dstType,
                                       typeConverter, rewriter);

    // Create a mask with bits set at [0, Count - 1].
    Value minusOne = createConstantAllBitsSet(loc, srcType, dstType, rewriter);
    Value maskShiftedByCount =
        rewriter.create<LLVM::ShlOp>(loc, dstType, minusOne, count);
    Value mask = rewriter.create<LLVM::XOrOp>(loc, dstType, maskShiftedByCount,
                                              minusOne);

    // Shift `Base` by `Offset` and apply the mask on it.
    Value shiftedBase =
        rewriter.create<LLVM::LShrOp>(loc, dstType, op.base(), offset);
    rewriter.replaceOpWithNewOp<LLVM::AndOp>(op, dstType, shiftedBase, mask);
    return success();
  }
};

class BranchConversionPattern : public SPIRVToLLVMConversion<spirv::BranchOp> {
public:
  using SPIRVToLLVMConversion<spirv::BranchOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::BranchOp branchOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<LLVM::BrOp>(branchOp, operands,
                                            branchOp.getTarget());
    return success();
  }
};

class BranchConditionalConversionPattern
    : public SPIRVToLLVMConversion<spirv::BranchConditionalOp> {
public:
  using SPIRVToLLVMConversion<
      spirv::BranchConditionalOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::BranchConditionalOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    // If branch weights exist, map them to 32-bit integer vector.
    ElementsAttr branchWeights = nullptr;
    if (auto weights = op.branch_weights()) {
      VectorType weightType = VectorType::get(2, rewriter.getI32Type());
      branchWeights =
          DenseElementsAttr::get(weightType, weights.getValue().getValue());
    }

    rewriter.replaceOpWithNewOp<LLVM::CondBrOp>(
        op, op.condition(), op.getTrueBlockArguments(),
        op.getFalseBlockArguments(), branchWeights, op.getTrueBlock(),
        op.getFalseBlock());
    return success();
  }
};

/// Converts `spv.CompositeExtract` to `llvm.extractvalue` if the container type
/// is an aggregate type (struct or array). Otherwise, converts to
/// `llvm.extractelement` that operates on vectors.
class CompositeExtractPattern
    : public SPIRVToLLVMConversion<spirv::CompositeExtractOp> {
public:
  using SPIRVToLLVMConversion<spirv::CompositeExtractOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::CompositeExtractOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto dstType = this->typeConverter.convertType(op.getType());
    if (!dstType)
      return failure();

    Type containerType = op.composite().getType();
    if (containerType.isa<VectorType>()) {
      Location loc = op.getLoc();
      IntegerAttr value = op.indices()[0].cast<IntegerAttr>();
      Value index = createI32ConstantOf(loc, rewriter, value.getInt());
      rewriter.replaceOpWithNewOp<LLVM::ExtractElementOp>(
          op, dstType, op.composite(), index);
      return success();
    }
    rewriter.replaceOpWithNewOp<LLVM::ExtractValueOp>(
        op, dstType, op.composite(), op.indices());
    return success();
  }
};

/// Converts `spv.CompositeInsert` to `llvm.insertvalue` if the container type
/// is an aggregate type (struct or array). Otherwise, converts to
/// `llvm.insertelement` that operates on vectors.
class CompositeInsertPattern
    : public SPIRVToLLVMConversion<spirv::CompositeInsertOp> {
public:
  using SPIRVToLLVMConversion<spirv::CompositeInsertOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::CompositeInsertOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto dstType = this->typeConverter.convertType(op.getType());
    if (!dstType)
      return failure();

    Type containerType = op.composite().getType();
    if (containerType.isa<VectorType>()) {
      Location loc = op.getLoc();
      IntegerAttr value = op.indices()[0].cast<IntegerAttr>();
      Value index = createI32ConstantOf(loc, rewriter, value.getInt());
      rewriter.replaceOpWithNewOp<LLVM::InsertElementOp>(
          op, dstType, op.composite(), op.object(), index);
      return success();
    }
    rewriter.replaceOpWithNewOp<LLVM::InsertValueOp>(
        op, dstType, op.composite(), op.object(), op.indices());
    return success();
  }
};

/// Converts SPIR-V operations that have straightforward LLVM equivalent
/// into LLVM dialect operations.
template <typename SPIRVOp, typename LLVMOp>
class DirectConversionPattern : public SPIRVToLLVMConversion<SPIRVOp> {
public:
  using SPIRVToLLVMConversion<SPIRVOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVOp operation, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto dstType = this->typeConverter.convertType(operation.getType());
    if (!dstType)
      return failure();
    rewriter.template replaceOpWithNewOp<LLVMOp>(operation, dstType, operands,
                                                 operation->getAttrs());
    return success();
  }
};

/// Converts `spv.ExecutionMode` into a global struct constant that holds
/// execution mode information.
class ExecutionModePattern
    : public SPIRVToLLVMConversion<spirv::ExecutionModeOp> {
public:
  using SPIRVToLLVMConversion<spirv::ExecutionModeOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::ExecutionModeOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    // First, create the global struct's name that would be associated with
    // this entry point's execution mode. We set it to be:
    //   __spv__{SPIR-V module name}_{function name}_execution_mode_info
    ModuleOp module = op->getParentOfType<ModuleOp>();
    std::string moduleName;
    if (module.getName().hasValue())
      moduleName = "_" + module.getName().getValue().str();
    else
      moduleName = "";
    std::string executionModeInfoName = llvm::formatv(
        "__spv_{0}_{1}_execution_mode_info", moduleName, op.fn().str());

    MLIRContext *context = rewriter.getContext();
    OpBuilder::InsertionGuard guard(rewriter);
    rewriter.setInsertionPointToStart(module.getBody());

    // Create a struct type, corresponding to the C struct below.
    // struct {
    //   int32_t executionMode;
    //   int32_t values[];          // optional values
    // };
    auto llvmI32Type = IntegerType::get(context, 32);
    SmallVector<Type, 2> fields;
    fields.push_back(llvmI32Type);
    ArrayAttr values = op.values();
    if (!values.empty()) {
      auto arrayType = LLVM::LLVMArrayType::get(llvmI32Type, values.size());
      fields.push_back(arrayType);
    }
    auto structType = LLVM::LLVMStructType::getLiteral(context, fields);

    // Create `llvm.mlir.global` with initializer region containing one block.
    auto global = rewriter.create<LLVM::GlobalOp>(
        UnknownLoc::get(context), structType, /*isConstant=*/true,
        LLVM::Linkage::External, executionModeInfoName, Attribute(),
        /*alignment=*/0);
    Location loc = global.getLoc();
    Region &region = global.getInitializerRegion();
    Block *block = rewriter.createBlock(&region);

    // Initialize the struct and set the execution mode value.
    rewriter.setInsertionPoint(block, block->begin());
    Value structValue = rewriter.create<LLVM::UndefOp>(loc, structType);
    IntegerAttr executionModeAttr = op.execution_modeAttr();
    Value executionMode =
        rewriter.create<LLVM::ConstantOp>(loc, llvmI32Type, executionModeAttr);
    structValue = rewriter.create<LLVM::InsertValueOp>(
        loc, structType, structValue, executionMode,
        ArrayAttr::get(context,
                       {rewriter.getIntegerAttr(rewriter.getI32Type(), 0)}));

    // Insert extra operands if they exist into execution mode info struct.
    for (unsigned i = 0, e = values.size(); i < e; ++i) {
      auto attr = values.getValue()[i];
      Value entry = rewriter.create<LLVM::ConstantOp>(loc, llvmI32Type, attr);
      structValue = rewriter.create<LLVM::InsertValueOp>(
          loc, structType, structValue, entry,
          ArrayAttr::get(context,
                         {rewriter.getIntegerAttr(rewriter.getI32Type(), 1),
                          rewriter.getIntegerAttr(rewriter.getI32Type(), i)}));
    }
    rewriter.create<LLVM::ReturnOp>(loc, ArrayRef<Value>({structValue}));
    rewriter.eraseOp(op);
    return success();
  }
};

/// Converts `spv.GlobalVariable` to `llvm.mlir.global`. Note that SPIR-V global
/// returns a pointer, whereas in LLVM dialect the global holds an actual value.
/// This difference is handled by `spv.mlir.addressof` and
/// `llvm.mlir.addressof`ops that both return a pointer.
class GlobalVariablePattern
    : public SPIRVToLLVMConversion<spirv::GlobalVariableOp> {
public:
  using SPIRVToLLVMConversion<spirv::GlobalVariableOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::GlobalVariableOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    // Currently, there is no support of initialization with a constant value in
    // SPIR-V dialect. Specialization constants are not considered as well.
    if (op.initializer())
      return failure();

    auto srcType = op.type().cast<spirv::PointerType>();
    auto dstType = typeConverter.convertType(srcType.getPointeeType());
    if (!dstType)
      return failure();

    // Limit conversion to the current invocation only or `StorageBuffer`
    // required by SPIR-V runner.
    // This is okay because multiple invocations are not supported yet.
    auto storageClass = srcType.getStorageClass();
    if (storageClass != spirv::StorageClass::Input &&
        storageClass != spirv::StorageClass::Private &&
        storageClass != spirv::StorageClass::Output &&
        storageClass != spirv::StorageClass::StorageBuffer) {
      return failure();
    }

    // LLVM dialect spec: "If the global value is a constant, storing into it is
    // not allowed.". This corresponds to SPIR-V 'Input' storage class that is
    // read-only.
    bool isConstant = storageClass == spirv::StorageClass::Input;
    // SPIR-V spec: "By default, functions and global variables are private to a
    // module and cannot be accessed by other modules. However, a module may be
    // written to export or import functions and global (module scope)
    // variables.". Therefore, map 'Private' storage class to private linkage,
    // 'Input' and 'Output' to external linkage.
    auto linkage = storageClass == spirv::StorageClass::Private
                       ? LLVM::Linkage::Private
                       : LLVM::Linkage::External;
    rewriter.replaceOpWithNewOp<LLVM::GlobalOp>(
        op, dstType, isConstant, linkage, op.sym_name(), Attribute(),
        /*alignment=*/0);
    return success();
  }
};

/// Converts SPIR-V cast ops that do not have straightforward LLVM
/// equivalent in LLVM dialect.
template <typename SPIRVOp, typename LLVMExtOp, typename LLVMTruncOp>
class IndirectCastPattern : public SPIRVToLLVMConversion<SPIRVOp> {
public:
  using SPIRVToLLVMConversion<SPIRVOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVOp operation, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    Type fromType = operation.operand().getType();
    Type toType = operation.getType();

    auto dstType = this->typeConverter.convertType(toType);
    if (!dstType)
      return failure();

    if (getBitWidth(fromType) < getBitWidth(toType)) {
      rewriter.template replaceOpWithNewOp<LLVMExtOp>(operation, dstType,
                                                      operands);
      return success();
    }
    if (getBitWidth(fromType) > getBitWidth(toType)) {
      rewriter.template replaceOpWithNewOp<LLVMTruncOp>(operation, dstType,
                                                        operands);
      return success();
    }
    return failure();
  }
};

class FunctionCallPattern
    : public SPIRVToLLVMConversion<spirv::FunctionCallOp> {
public:
  using SPIRVToLLVMConversion<spirv::FunctionCallOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::FunctionCallOp callOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    if (callOp.getNumResults() == 0) {
      rewriter.replaceOpWithNewOp<LLVM::CallOp>(callOp, llvm::None, operands,
                                                callOp->getAttrs());
      return success();
    }

    // Function returns a single result.
    auto dstType = typeConverter.convertType(callOp.getType(0));
    rewriter.replaceOpWithNewOp<LLVM::CallOp>(callOp, dstType, operands,
                                              callOp->getAttrs());
    return success();
  }
};

/// Converts SPIR-V floating-point comparisons to llvm.fcmp "predicate"
template <typename SPIRVOp, LLVM::FCmpPredicate predicate>
class FComparePattern : public SPIRVToLLVMConversion<SPIRVOp> {
public:
  using SPIRVToLLVMConversion<SPIRVOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVOp operation, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    auto dstType = this->typeConverter.convertType(operation.getType());
    if (!dstType)
      return failure();

    rewriter.template replaceOpWithNewOp<LLVM::FCmpOp>(
        operation, dstType, predicate, operation.operand1(),
        operation.operand2());
    return success();
  }
};

/// Converts SPIR-V integer comparisons to llvm.icmp "predicate"
template <typename SPIRVOp, LLVM::ICmpPredicate predicate>
class IComparePattern : public SPIRVToLLVMConversion<SPIRVOp> {
public:
  using SPIRVToLLVMConversion<SPIRVOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVOp operation, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    auto dstType = this->typeConverter.convertType(operation.getType());
    if (!dstType)
      return failure();

    rewriter.template replaceOpWithNewOp<LLVM::ICmpOp>(
        operation, dstType, predicate, operation.operand1(),
        operation.operand2());
    return success();
  }
};

class InverseSqrtPattern
    : public SPIRVToLLVMConversion<spirv::GLSLInverseSqrtOp> {
public:
  using SPIRVToLLVMConversion<spirv::GLSLInverseSqrtOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::GLSLInverseSqrtOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto srcType = op.getType();
    auto dstType = typeConverter.convertType(srcType);
    if (!dstType)
      return failure();

    Location loc = op.getLoc();
    Value one = createFPConstant(loc, srcType, dstType, rewriter, 1.0);
    Value sqrt = rewriter.create<LLVM::SqrtOp>(loc, dstType, op.operand());
    rewriter.replaceOpWithNewOp<LLVM::FDivOp>(op, dstType, one, sqrt);
    return success();
  }
};

/// Converts `spv.Load` and `spv.Store` to LLVM dialect.
template <typename SPIRVop>
class LoadStorePattern : public SPIRVToLLVMConversion<SPIRVop> {
public:
  using SPIRVToLLVMConversion<SPIRVop>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVop op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    if (!op.memory_access().hasValue()) {
      return replaceWithLoadOrStore(
          op, rewriter, this->typeConverter, /*alignment=*/0,
          /*isVolatile=*/false, /*isNonTemporal=*/false);
    }
    auto memoryAccess = op.memory_access().getValue();
    switch (memoryAccess) {
    case spirv::MemoryAccess::Aligned:
    case spirv::MemoryAccess::None:
    case spirv::MemoryAccess::Nontemporal:
    case spirv::MemoryAccess::Volatile: {
      unsigned alignment =
          memoryAccess == spirv::MemoryAccess::Aligned ? *op.alignment() : 0;
      bool isNonTemporal = memoryAccess == spirv::MemoryAccess::Nontemporal;
      bool isVolatile = memoryAccess == spirv::MemoryAccess::Volatile;
      return replaceWithLoadOrStore(op, rewriter, this->typeConverter,
                                    alignment, isVolatile, isNonTemporal);
    }
    default:
      // There is no support of other memory access attributes.
      return failure();
    }
  }
};

/// Converts `spv.Not` and `spv.LogicalNot` into LLVM dialect.
template <typename SPIRVOp>
class NotPattern : public SPIRVToLLVMConversion<SPIRVOp> {
public:
  using SPIRVToLLVMConversion<SPIRVOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVOp notOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    auto srcType = notOp.getType();
    auto dstType = this->typeConverter.convertType(srcType);
    if (!dstType)
      return failure();

    Location loc = notOp.getLoc();
    IntegerAttr minusOne = minusOneIntegerAttribute(srcType, rewriter);
    auto mask = srcType.template isa<VectorType>()
                    ? rewriter.create<LLVM::ConstantOp>(
                          loc, dstType,
                          SplatElementsAttr::get(
                              srcType.template cast<VectorType>(), minusOne))
                    : rewriter.create<LLVM::ConstantOp>(loc, dstType, minusOne);
    rewriter.template replaceOpWithNewOp<LLVM::XOrOp>(notOp, dstType,
                                                      notOp.operand(), mask);
    return success();
  }
};

/// A template pattern that erases the given `SPIRVOp`.
template <typename SPIRVOp>
class ErasePattern : public SPIRVToLLVMConversion<SPIRVOp> {
public:
  using SPIRVToLLVMConversion<SPIRVOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.eraseOp(op);
    return success();
  }
};

class ReturnPattern : public SPIRVToLLVMConversion<spirv::ReturnOp> {
public:
  using SPIRVToLLVMConversion<spirv::ReturnOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::ReturnOp returnOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<LLVM::ReturnOp>(returnOp, ArrayRef<Type>(),
                                                ArrayRef<Value>());
    return success();
  }
};

class ReturnValuePattern : public SPIRVToLLVMConversion<spirv::ReturnValueOp> {
public:
  using SPIRVToLLVMConversion<spirv::ReturnValueOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::ReturnValueOp returnValueOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<LLVM::ReturnOp>(returnValueOp, ArrayRef<Type>(),
                                                operands);
    return success();
  }
};

/// Converts `spv.mlir.loop` to LLVM dialect. All blocks within selection should
/// be reachable for conversion to succeed. The structure of the loop in LLVM
/// dialect will be the following:
///
///      +------------------------------------+
///      | <code before spv.mlir.loop>        |
///      | llvm.br ^header                    |
///      +------------------------------------+
///                           |
///   +----------------+      |
///   |                |      |
///   |                V      V
///   |  +------------------------------------+
///   |  | ^header:                           |
///   |  |   <header code>                    |
///   |  |   llvm.cond_br %cond, ^body, ^exit |
///   |  +------------------------------------+
///   |                    |
///   |                    |----------------------+
///   |                    |                      |
///   |                    V                      |
///   |  +------------------------------------+   |
///   |  | ^body:                             |   |
///   |  |   <body code>                      |   |
///   |  |   llvm.br ^continue                |   |
///   |  +------------------------------------+   |
///   |                    |                      |
///   |                    V                      |
///   |  +------------------------------------+   |
///   |  | ^continue:                         |   |
///   |  |   <continue code>                  |   |
///   |  |   llvm.br ^header                  |   |
///   |  +------------------------------------+   |
///   |               |                           |
///   +---------------+    +----------------------+
///                        |
///                        V
///      +------------------------------------+
///      | ^exit:                             |
///      |   llvm.br ^remaining               |
///      +------------------------------------+
///                        |
///                        V
///      +------------------------------------+
///      | ^remaining:                        |
///      |   <code after spv.mlir.loop>       |
///      +------------------------------------+
///
class LoopPattern : public SPIRVToLLVMConversion<spirv::LoopOp> {
public:
  using SPIRVToLLVMConversion<spirv::LoopOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::LoopOp loopOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    // There is no support of loop control at the moment.
    if (loopOp.loop_control() != spirv::LoopControl::None)
      return failure();

    Location loc = loopOp.getLoc();

    // Split the current block after `spv.mlir.loop`. The remaining ops will be
    // used in `endBlock`.
    Block *currentBlock = rewriter.getBlock();
    auto position = Block::iterator(loopOp);
    Block *endBlock = rewriter.splitBlock(currentBlock, position);

    // Remove entry block and create a branch in the current block going to the
    // header block.
    Block *entryBlock = loopOp.getEntryBlock();
    assert(entryBlock->getOperations().size() == 1);
    auto brOp = dyn_cast<spirv::BranchOp>(entryBlock->getOperations().front());
    if (!brOp)
      return failure();
    Block *headerBlock = loopOp.getHeaderBlock();
    rewriter.setInsertionPointToEnd(currentBlock);
    rewriter.create<LLVM::BrOp>(loc, brOp.getBlockArguments(), headerBlock);
    rewriter.eraseBlock(entryBlock);

    // Branch from merge block to end block.
    Block *mergeBlock = loopOp.getMergeBlock();
    Operation *terminator = mergeBlock->getTerminator();
    ValueRange terminatorOperands = terminator->getOperands();
    rewriter.setInsertionPointToEnd(mergeBlock);
    rewriter.create<LLVM::BrOp>(loc, terminatorOperands, endBlock);

    rewriter.inlineRegionBefore(loopOp.body(), endBlock);
    rewriter.replaceOp(loopOp, endBlock->getArguments());
    return success();
  }
};

/// Converts `spv.mlir.selection` with `spv.BranchConditional` in its header
/// block. All blocks within selection should be reachable for conversion to
/// succeed.
class SelectionPattern : public SPIRVToLLVMConversion<spirv::SelectionOp> {
public:
  using SPIRVToLLVMConversion<spirv::SelectionOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::SelectionOp op, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    // There is no support for `Flatten` or `DontFlatten` selection control at
    // the moment. This are just compiler hints and can be performed during the
    // optimization passes.
    if (op.selection_control() != spirv::SelectionControl::None)
      return failure();

    // `spv.mlir.selection` should have at least two blocks: one selection
    // header block and one merge block. If no blocks are present, or control
    // flow branches straight to merge block (two blocks are present), the op is
    // redundant and it is erased.
    if (op.body().getBlocks().size() <= 2) {
      rewriter.eraseOp(op);
      return success();
    }

    Location loc = op.getLoc();

    // Split the current block after `spv.mlir.selection`. The remaining ops
    // will be used in `continueBlock`.
    auto *currentBlock = rewriter.getInsertionBlock();
    rewriter.setInsertionPointAfter(op);
    auto position = rewriter.getInsertionPoint();
    auto *continueBlock = rewriter.splitBlock(currentBlock, position);

    // Extract conditional branch information from the header block. By SPIR-V
    // dialect spec, it should contain `spv.BranchConditional` or `spv.Switch`
    // op. Note that `spv.Switch op` is not supported at the moment in the
    // SPIR-V dialect. Remove this block when finished.
    auto *headerBlock = op.getHeaderBlock();
    assert(headerBlock->getOperations().size() == 1);
    auto condBrOp = dyn_cast<spirv::BranchConditionalOp>(
        headerBlock->getOperations().front());
    if (!condBrOp)
      return failure();
    rewriter.eraseBlock(headerBlock);

    // Branch from merge block to continue block.
    auto *mergeBlock = op.getMergeBlock();
    Operation *terminator = mergeBlock->getTerminator();
    ValueRange terminatorOperands = terminator->getOperands();
    rewriter.setInsertionPointToEnd(mergeBlock);
    rewriter.create<LLVM::BrOp>(loc, terminatorOperands, continueBlock);

    // Link current block to `true` and `false` blocks within the selection.
    Block *trueBlock = condBrOp.getTrueBlock();
    Block *falseBlock = condBrOp.getFalseBlock();
    rewriter.setInsertionPointToEnd(currentBlock);
    rewriter.create<LLVM::CondBrOp>(loc, condBrOp.condition(), trueBlock,
                                    condBrOp.trueTargetOperands(), falseBlock,
                                    condBrOp.falseTargetOperands());

    rewriter.inlineRegionBefore(op.body(), continueBlock);
    rewriter.replaceOp(op, continueBlock->getArguments());
    return success();
  }
};

/// Converts SPIR-V shift ops to LLVM shift ops. Since LLVM dialect
/// puts a restriction on `Shift` and `Base` to have the same bit width,
/// `Shift` is zero or sign extended to match this specification. Cases when
/// `Shift` bit width > `Base` bit width are considered to be illegal.
template <typename SPIRVOp, typename LLVMOp>
class ShiftPattern : public SPIRVToLLVMConversion<SPIRVOp> {
public:
  using SPIRVToLLVMConversion<SPIRVOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(SPIRVOp operation, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    auto dstType = this->typeConverter.convertType(operation.getType());
    if (!dstType)
      return failure();

    Type op1Type = operation.operand1().getType();
    Type op2Type = operation.operand2().getType();

    if (op1Type == op2Type) {
      rewriter.template replaceOpWithNewOp<LLVMOp>(operation, dstType,
                                                   operands);
      return success();
    }

    Location loc = operation.getLoc();
    Value extended;
    if (isUnsignedIntegerOrVector(op2Type)) {
      extended = rewriter.template create<LLVM::ZExtOp>(loc, dstType,
                                                        operation.operand2());
    } else {
      extended = rewriter.template create<LLVM::SExtOp>(loc, dstType,
                                                        operation.operand2());
    }
    Value result = rewriter.template create<LLVMOp>(
        loc, dstType, operation.operand1(), extended);
    rewriter.replaceOp(operation, result);
    return success();
  }
};

class TanPattern : public SPIRVToLLVMConversion<spirv::GLSLTanOp> {
public:
  using SPIRVToLLVMConversion<spirv::GLSLTanOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::GLSLTanOp tanOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto dstType = typeConverter.convertType(tanOp.getType());
    if (!dstType)
      return failure();

    Location loc = tanOp.getLoc();
    Value sin = rewriter.create<LLVM::SinOp>(loc, dstType, tanOp.operand());
    Value cos = rewriter.create<LLVM::CosOp>(loc, dstType, tanOp.operand());
    rewriter.replaceOpWithNewOp<LLVM::FDivOp>(tanOp, dstType, sin, cos);
    return success();
  }
};

/// Convert `spv.Tanh` to
///
///   exp(2x) - 1
///   -----------
///   exp(2x) + 1
///
class TanhPattern : public SPIRVToLLVMConversion<spirv::GLSLTanhOp> {
public:
  using SPIRVToLLVMConversion<spirv::GLSLTanhOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::GLSLTanhOp tanhOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto srcType = tanhOp.getType();
    auto dstType = typeConverter.convertType(srcType);
    if (!dstType)
      return failure();

    Location loc = tanhOp.getLoc();
    Value two = createFPConstant(loc, srcType, dstType, rewriter, 2.0);
    Value multiplied =
        rewriter.create<LLVM::FMulOp>(loc, dstType, two, tanhOp.operand());
    Value exponential = rewriter.create<LLVM::ExpOp>(loc, dstType, multiplied);
    Value one = createFPConstant(loc, srcType, dstType, rewriter, 1.0);
    Value numerator =
        rewriter.create<LLVM::FSubOp>(loc, dstType, exponential, one);
    Value denominator =
        rewriter.create<LLVM::FAddOp>(loc, dstType, exponential, one);
    rewriter.replaceOpWithNewOp<LLVM::FDivOp>(tanhOp, dstType, numerator,
                                              denominator);
    return success();
  }
};

class VariablePattern : public SPIRVToLLVMConversion<spirv::VariableOp> {
public:
  using SPIRVToLLVMConversion<spirv::VariableOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::VariableOp varOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {
    auto srcType = varOp.getType();
    // Initialization is supported for scalars and vectors only.
    auto pointerTo = srcType.cast<spirv::PointerType>().getPointeeType();
    auto init = varOp.initializer();
    if (init && !pointerTo.isIntOrFloat() && !pointerTo.isa<VectorType>())
      return failure();

    auto dstType = typeConverter.convertType(srcType);
    if (!dstType)
      return failure();

    Location loc = varOp.getLoc();
    Value size = createI32ConstantOf(loc, rewriter, 1);
    if (!init) {
      rewriter.replaceOpWithNewOp<LLVM::AllocaOp>(varOp, dstType, size);
      return success();
    }
    Value allocated = rewriter.create<LLVM::AllocaOp>(loc, dstType, size);
    rewriter.create<LLVM::StoreOp>(loc, init, allocated);
    rewriter.replaceOp(varOp, allocated);
    return success();
  }
};

//===----------------------------------------------------------------------===//
// FuncOp conversion
//===----------------------------------------------------------------------===//

class FuncConversionPattern : public SPIRVToLLVMConversion<spirv::FuncOp> {
public:
  using SPIRVToLLVMConversion<spirv::FuncOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::FuncOp funcOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    // Convert function signature. At the moment LLVMType converter is enough
    // for currently supported types.
    auto funcType = funcOp.getType();
    TypeConverter::SignatureConversion signatureConverter(
        funcType.getNumInputs());
    auto llvmType = typeConverter.convertFunctionSignature(
        funcOp.getType(), /*isVariadic=*/false, signatureConverter);
    if (!llvmType)
      return failure();

    // Create a new `LLVMFuncOp`
    Location loc = funcOp.getLoc();
    StringRef name = funcOp.getName();
    auto newFuncOp = rewriter.create<LLVM::LLVMFuncOp>(loc, name, llvmType);

    // Convert SPIR-V Function Control to equivalent LLVM function attribute
    MLIRContext *context = funcOp.getContext();
    switch (funcOp.function_control()) {
#define DISPATCH(functionControl, llvmAttr)                                    \
  case functionControl:                                                        \
    newFuncOp->setAttr("passthrough", ArrayAttr::get(context, {llvmAttr}));    \
    break;

      DISPATCH(spirv::FunctionControl::Inline,
               StringAttr::get(context, "alwaysinline"));
      DISPATCH(spirv::FunctionControl::DontInline,
               StringAttr::get(context, "noinline"));
      DISPATCH(spirv::FunctionControl::Pure,
               StringAttr::get(context, "readonly"));
      DISPATCH(spirv::FunctionControl::Const,
               StringAttr::get(context, "readnone"));

#undef DISPATCH

    // Default: if `spirv::FunctionControl::None`, then no attributes are
    // needed.
    default:
      break;
    }

    rewriter.inlineRegionBefore(funcOp.getBody(), newFuncOp.getBody(),
                                newFuncOp.end());
    if (failed(rewriter.convertRegionTypes(&newFuncOp.getBody(), typeConverter,
                                           &signatureConverter))) {
      return failure();
    }
    rewriter.eraseOp(funcOp);
    return success();
  }
};

//===----------------------------------------------------------------------===//
// ModuleOp conversion
//===----------------------------------------------------------------------===//

class ModuleConversionPattern : public SPIRVToLLVMConversion<spirv::ModuleOp> {
public:
  using SPIRVToLLVMConversion<spirv::ModuleOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::ModuleOp spvModuleOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    auto newModuleOp =
        rewriter.create<ModuleOp>(spvModuleOp.getLoc(), spvModuleOp.getName());
    rewriter.inlineRegionBefore(spvModuleOp.body(), newModuleOp.getBody());

    // Remove the terminator block that was automatically added by builder
    rewriter.eraseBlock(&newModuleOp.getBodyRegion().back());
    rewriter.eraseOp(spvModuleOp);
    return success();
  }
};

class ModuleEndConversionPattern
    : public SPIRVToLLVMConversion<spirv::ModuleEndOp> {
public:
  using SPIRVToLLVMConversion<spirv::ModuleEndOp>::SPIRVToLLVMConversion;

  LogicalResult
  matchAndRewrite(spirv::ModuleEndOp moduleEndOp, ArrayRef<Value> operands,
                  ConversionPatternRewriter &rewriter) const override {

    rewriter.eraseOp(moduleEndOp);
    return success();
  }
};

} // namespace

//===----------------------------------------------------------------------===//
// Pattern population
//===----------------------------------------------------------------------===//

void mlir::populateSPIRVToLLVMTypeConversion(LLVMTypeConverter &typeConverter) {
  typeConverter.addConversion([&](spirv::ArrayType type) {
    return convertArrayType(type, typeConverter);
  });
  typeConverter.addConversion([&](spirv::PointerType type) {
    return convertPointerType(type, typeConverter);
  });
  typeConverter.addConversion([&](spirv::RuntimeArrayType type) {
    return convertRuntimeArrayType(type, typeConverter);
  });
  typeConverter.addConversion([&](spirv::StructType type) {
    return convertStructType(type, typeConverter);
  });
}

void mlir::populateSPIRVToLLVMConversionPatterns(
    LLVMTypeConverter &typeConverter, RewritePatternSet &patterns) {
  patterns.add<
      // Arithmetic ops
      DirectConversionPattern<spirv::IAddOp, LLVM::AddOp>,
      DirectConversionPattern<spirv::IMulOp, LLVM::MulOp>,
      DirectConversionPattern<spirv::ISubOp, LLVM::SubOp>,
      DirectConversionPattern<spirv::FAddOp, LLVM::FAddOp>,
      DirectConversionPattern<spirv::FDivOp, LLVM::FDivOp>,
      DirectConversionPattern<spirv::FMulOp, LLVM::FMulOp>,
      DirectConversionPattern<spirv::FNegateOp, LLVM::FNegOp>,
      DirectConversionPattern<spirv::FRemOp, LLVM::FRemOp>,
      DirectConversionPattern<spirv::FSubOp, LLVM::FSubOp>,
      DirectConversionPattern<spirv::SDivOp, LLVM::SDivOp>,
      DirectConversionPattern<spirv::SRemOp, LLVM::SRemOp>,
      DirectConversionPattern<spirv::UDivOp, LLVM::UDivOp>,
      DirectConversionPattern<spirv::UModOp, LLVM::URemOp>,

      // Bitwise ops
      BitFieldInsertPattern, BitFieldUExtractPattern, BitFieldSExtractPattern,
      DirectConversionPattern<spirv::BitCountOp, LLVM::CtPopOp>,
      DirectConversionPattern<spirv::BitReverseOp, LLVM::BitReverseOp>,
      DirectConversionPattern<spirv::BitwiseAndOp, LLVM::AndOp>,
      DirectConversionPattern<spirv::BitwiseOrOp, LLVM::OrOp>,
      DirectConversionPattern<spirv::BitwiseXorOp, LLVM::XOrOp>,
      NotPattern<spirv::NotOp>,

      // Cast ops
      DirectConversionPattern<spirv::BitcastOp, LLVM::BitcastOp>,
      DirectConversionPattern<spirv::ConvertFToSOp, LLVM::FPToSIOp>,
      DirectConversionPattern<spirv::ConvertFToUOp, LLVM::FPToUIOp>,
      DirectConversionPattern<spirv::ConvertSToFOp, LLVM::SIToFPOp>,
      DirectConversionPattern<spirv::ConvertUToFOp, LLVM::UIToFPOp>,
      IndirectCastPattern<spirv::FConvertOp, LLVM::FPExtOp, LLVM::FPTruncOp>,
      IndirectCastPattern<spirv::SConvertOp, LLVM::SExtOp, LLVM::TruncOp>,
      IndirectCastPattern<spirv::UConvertOp, LLVM::ZExtOp, LLVM::TruncOp>,

      // Comparison ops
      IComparePattern<spirv::IEqualOp, LLVM::ICmpPredicate::eq>,
      IComparePattern<spirv::INotEqualOp, LLVM::ICmpPredicate::ne>,
      FComparePattern<spirv::FOrdEqualOp, LLVM::FCmpPredicate::oeq>,
      FComparePattern<spirv::FOrdGreaterThanOp, LLVM::FCmpPredicate::ogt>,
      FComparePattern<spirv::FOrdGreaterThanEqualOp, LLVM::FCmpPredicate::oge>,
      FComparePattern<spirv::FOrdLessThanEqualOp, LLVM::FCmpPredicate::ole>,
      FComparePattern<spirv::FOrdLessThanOp, LLVM::FCmpPredicate::olt>,
      FComparePattern<spirv::FOrdNotEqualOp, LLVM::FCmpPredicate::one>,
      FComparePattern<spirv::FUnordEqualOp, LLVM::FCmpPredicate::ueq>,
      FComparePattern<spirv::FUnordGreaterThanOp, LLVM::FCmpPredicate::ugt>,
      FComparePattern<spirv::FUnordGreaterThanEqualOp,
                      LLVM::FCmpPredicate::uge>,
      FComparePattern<spirv::FUnordLessThanEqualOp, LLVM::FCmpPredicate::ule>,
      FComparePattern<spirv::FUnordLessThanOp, LLVM::FCmpPredicate::ult>,
      FComparePattern<spirv::FUnordNotEqualOp, LLVM::FCmpPredicate::une>,
      IComparePattern<spirv::SGreaterThanOp, LLVM::ICmpPredicate::sgt>,
      IComparePattern<spirv::SGreaterThanEqualOp, LLVM::ICmpPredicate::sge>,
      IComparePattern<spirv::SLessThanEqualOp, LLVM::ICmpPredicate::sle>,
      IComparePattern<spirv::SLessThanOp, LLVM::ICmpPredicate::slt>,
      IComparePattern<spirv::UGreaterThanOp, LLVM::ICmpPredicate::ugt>,
      IComparePattern<spirv::UGreaterThanEqualOp, LLVM::ICmpPredicate::uge>,
      IComparePattern<spirv::ULessThanEqualOp, LLVM::ICmpPredicate::ule>,
      IComparePattern<spirv::ULessThanOp, LLVM::ICmpPredicate::ult>,

      // Constant op
      ConstantScalarAndVectorPattern,

      // Control Flow ops
      BranchConversionPattern, BranchConditionalConversionPattern,
      FunctionCallPattern, LoopPattern, SelectionPattern,
      ErasePattern<spirv::MergeOp>,

      // Entry points and execution mode are handled separately.
      ErasePattern<spirv::EntryPointOp>, ExecutionModePattern,

      // GLSL extended instruction set ops
      DirectConversionPattern<spirv::GLSLCeilOp, LLVM::FCeilOp>,
      DirectConversionPattern<spirv::GLSLCosOp, LLVM::CosOp>,
      DirectConversionPattern<spirv::GLSLExpOp, LLVM::ExpOp>,
      DirectConversionPattern<spirv::GLSLFAbsOp, LLVM::FAbsOp>,
      DirectConversionPattern<spirv::GLSLFloorOp, LLVM::FFloorOp>,
      DirectConversionPattern<spirv::GLSLFMaxOp, LLVM::MaxNumOp>,
      DirectConversionPattern<spirv::GLSLFMinOp, LLVM::MinNumOp>,
      DirectConversionPattern<spirv::GLSLLogOp, LLVM::LogOp>,
      DirectConversionPattern<spirv::GLSLSinOp, LLVM::SinOp>,
      DirectConversionPattern<spirv::GLSLSMaxOp, LLVM::SMaxOp>,
      DirectConversionPattern<spirv::GLSLSMinOp, LLVM::SMinOp>,
      DirectConversionPattern<spirv::GLSLSqrtOp, LLVM::SqrtOp>,
      InverseSqrtPattern, TanPattern, TanhPattern,

      // Logical ops
      DirectConversionPattern<spirv::LogicalAndOp, LLVM::AndOp>,
      DirectConversionPattern<spirv::LogicalOrOp, LLVM::OrOp>,
      IComparePattern<spirv::LogicalEqualOp, LLVM::ICmpPredicate::eq>,
      IComparePattern<spirv::LogicalNotEqualOp, LLVM::ICmpPredicate::ne>,
      NotPattern<spirv::LogicalNotOp>,

      // Memory ops
      AccessChainPattern, AddressOfPattern, GlobalVariablePattern,
      LoadStorePattern<spirv::LoadOp>, LoadStorePattern<spirv::StoreOp>,
      VariablePattern,

      // Miscellaneous ops
      CompositeExtractPattern, CompositeInsertPattern,
      DirectConversionPattern<spirv::SelectOp, LLVM::SelectOp>,
      DirectConversionPattern<spirv::UndefOp, LLVM::UndefOp>,

      // Shift ops
      ShiftPattern<spirv::ShiftRightArithmeticOp, LLVM::AShrOp>,
      ShiftPattern<spirv::ShiftRightLogicalOp, LLVM::LShrOp>,
      ShiftPattern<spirv::ShiftLeftLogicalOp, LLVM::ShlOp>,

      // Return ops
      ReturnPattern, ReturnValuePattern>(patterns.getContext(), typeConverter);
}

void mlir::populateSPIRVToLLVMFunctionConversionPatterns(
    LLVMTypeConverter &typeConverter, RewritePatternSet &patterns) {
  patterns.add<FuncConversionPattern>(patterns.getContext(), typeConverter);
}

void mlir::populateSPIRVToLLVMModuleConversionPatterns(
    LLVMTypeConverter &typeConverter, RewritePatternSet &patterns) {
  patterns.add<ModuleConversionPattern, ModuleEndConversionPattern>(
      patterns.getContext(), typeConverter);
}

//===----------------------------------------------------------------------===//
// Pre-conversion hooks
//===----------------------------------------------------------------------===//

/// Hook for descriptor set and binding number encoding.
static constexpr StringRef kBinding = "binding";
static constexpr StringRef kDescriptorSet = "descriptor_set";
void mlir::encodeBindAttribute(ModuleOp module) {
  auto spvModules = module.getOps<spirv::ModuleOp>();
  for (auto spvModule : spvModules) {
    spvModule.walk([&](spirv::GlobalVariableOp op) {
      IntegerAttr descriptorSet =
          op->getAttrOfType<IntegerAttr>(kDescriptorSet);
      IntegerAttr binding = op->getAttrOfType<IntegerAttr>(kBinding);
      // For every global variable in the module, get the ones with descriptor
      // set and binding numbers.
      if (descriptorSet && binding) {
        // Encode these numbers into the variable's symbolic name. If the
        // SPIR-V module has a name, add it at the beginning.
        auto moduleAndName = spvModule.getName().hasValue()
                                 ? spvModule.getName().getValue().str() + "_" +
                                       op.sym_name().str()
                                 : op.sym_name().str();
        std::string name =
            llvm::formatv("{0}_descriptor_set{1}_binding{2}", moduleAndName,
                          std::to_string(descriptorSet.getInt()),
                          std::to_string(binding.getInt()));

        // Replace all symbol uses and set the new symbol name. Finally, remove
        // descriptor set and binding attributes.
        if (failed(SymbolTable::replaceAllSymbolUses(op, name, spvModule)))
          op.emitError("unable to replace all symbol uses for ") << name;
        SymbolTable::setSymbolName(op, name);
        op->removeAttr(kDescriptorSet);
        op->removeAttr(kBinding);
      }
    });
  }
}